Questions, answered plainly.

With a free operations audit. Fill out the quote form or email hello@crowork.ai. You describe how the work happens; we map where AI can remove, automate, or improve something and hand that back in plain language with an expected ROI. No obligation, no charge. If there is no clear win, we say so.

A description of the problem or workflow you want automated. No spec, no brief. The quote form asks the right questions to scope it together in the first conversation.

Depends on scope — we give you a concrete timeline after the audit, not a generic one here. Most single-workflow builds ship in 5–10 business days. Complex multi-system agents can take 2–4 weeks. We commit to a real date and meet it.

Both. A single pipeline, a custom agent, or a longer retainer where we run and improve the systems we build. Most clients start with one concrete problem and expand from there.

AI models: Anthropic Claude, OpenAI/Codex, and Z.AI/GLM — selected per task based on capability and cost. Infrastructure: Fly.io for deployment, Supabase (PostgreSQL) for structured data and auth. Integration layer: n8n for workflow automation, Telegram MTProto and WhatsApp Business API for messaging. Google Workspace for email and calendar flows. TypeScript and Python throughout. We pick the right tool per job; we do not lock clients into one platform.

BYOK means Bring Your Own Keys — you supply your own Anthropic, OpenAI, or other API credentials and the system uses your account's quota directly. The alternative is managed keys, where we hold credentials on your behalf. Both models are supported. BYOK gives you full cost visibility and keeps all AI provider billing under your own account. Managed keys are simpler to start with. Either way, keys are stored encrypted at rest and never in plaintext — a mistake we have seen in the wild too many times.

The agents we build log every action, classify failures, and write their own diagnostic notes. When a repeating failure pattern emerges — an API contract change, a data-format drift, a flaky integration — the system surfaces it with a proposed fix rather than silently degrading. Over time, it patches known failure modes without human intervention. We run our own operations this way; it is the same stack we hand to clients.

Usually not. We prefer to wire AI into the tools your team already uses — email, CRM, spreadsheets, the chat platform you live in. The best automation is the kind nobody has to learn. Adoption is where most projects die, so we design around it.

Permissions are scoped to the role — we take only the access a job actually needs. Workers run with a full action log, so you can audit exactly what was touched and when. We do not repurpose your data. What we build is yours. For anything sensitive, we agree on handling rules in writing during the audit, before any access is granted.

Yes. Systems are deployed on Fly.io with EU region targeting and backed by Supabase with EU data residency. Data is encrypted at rest. We do not store API keys in plaintext — credentials go through environment-variable injection or a secrets manager, never a config file. If you have specific DPA requirements, we address them during scoping.

Supervised by design. Every action is logged, edge cases are held for a human rather than guessed, and you can pause, correct, or roll back at any point. We consider an unsupervised agent acting on your operation a liability, not a feature.

Our reliability tooling exists specifically to solve this. Before anything is stored as a fact, it is classified — a question is not a fact, a hypothetical is not a fact, a third-party statement is not a first-person assertion. A trust model means a single mention is not treated as certainty; facts earn confidence through repetition and lose it through silence. We would rather a worker say "I do not know" than confidently invent an answer.

Flat and fixed. A single price for a defined build scope, and a flat monthly rate to run and maintain it. We never bill per API token, per call, or per message — your cost does not move with usage. You know the number before you sign.

You do. We hand over the working system and its documentation. We are not interested in a black box that only we can maintain or that locks you in.

Then we tell you during the free audit, before you have spent anything. We would rather lose a project than sell automation that does not earn its place. Trust is worth more to us than one engagement.

We run our own operations on the same kind of AI systems we build for clients, and we open-source the reliability tooling underneath them. We are not a traditional consultancy that added an AI line to its menu. Building and running these systems is the whole company.

No account managers, no project managers, no juniors learning on your budget. You speak directly with the person building. We are faster, leaner, and already run the stack we build for you. It is not theoretical.

From individual operators to small and mid-sized teams. The common thread is a real, repetitive operational pain — a report rebuilt by hand every day, an inbox burying commitments, a monitoring seat spent on vigilance. If that sounds like you, the audit will tell us quickly whether there is a fit.